Rebellions, a South Korean fabless AI chip startup, said today that it has raised $124 million (165 billion KRW) in a Series B round of investment for the development of its third AI chip, Rebel. The business will also utilize the fresh money, which was oversubscribed with an initial target of $90 million, to increase production of its data center-focused microprocessor, Atom, as well as to hire.
Rebellions’ CFO, Sungkyue Shin, told Eltrys in an exclusive interview that the three-year-old firm is valued at around $658 million (880 billion KRW) post-money. This recent financial injection takes Rebellions’ total raised to over $210 million since its debut in 2020.
KT, the South Korean telecom behemoth, led the newest round as a strategic investor. Previous supporters Temasek Pavilion Capital and Korea Development Bank, as well as new investors Korelya Capital and DG Daiwa Ventures, took part.
Rebellions’ fundraising comes at a critical juncture in the semiconductor industry, notably in the development and use of AI circuits.
Nvidia is the market leader in AI chips, and its brand is linked with the current AI boom in the technological industry. Many people have noticed how Nvidia has grown in part because of the moat built around an ecosystem of hardware and software. However, the remainder of the field is not out of the running. Data processing and its high costs remain important difficulties in AI applications; thus, the quest for novel innovations to better them continues.
Developments are occurring on numerous fronts. Big IT heavyweights like Google, Amazon, Apple, and Microsoft create or own chips for incorporating AI into their goods and services. OpenAI CEO Sam Altman allegedly visited South Korea last week to meet with the country’s semiconductor industry titans, Samsung and SK Hynix. Beyond that, OpenAI is rumored to be funding billions of dollars to establish chip fabrication factories to manufacture its own AI processors. Beyond rebellions, there are a lot of businesses introducing innovative methods to speed up processing while enhancing efficiency.
Team up with Samsung.
This fundraiser, which has been reported for months, follows earlier initiatives by the firm. Last October, Rebellions announced that it would collaborate with Samsung Electronics to produce its newest Rebel chip, expanding on a collaboration that began with its Atom processors. The two businesses want to conclude Rebel development by the end of this year and begin mass production in 2025, according to Shin, who also stated that the next-generation AI chip would target the generative AI industry, which includes large language models (LLMs) and hyperscalers.
According to Shin, Rebel will employ Samsung Electronics’ 4-nanometer fabrication process, and its AI chip will be integrated with Samsung’s advanced memory chip technology, HBM3E, which is designed to handle high-bandwidth memory and is used for generating and managing big language models. Rebellions’ unique selling advantage is that its technology and solutions are more versatile than proprietary AI chips, allowing them to accommodate a wide range of generative AI models that require AI accelerators.
The company’s CFO stated that Rebellions will collaborate with Samsung on everything from co-development and chip design to mass manufacture of Rebel. There is a second incentive to Samsung’s work here: In addition to its semiconductor efforts, South Korea’s leading memory chip manufacturer, Samsung, has been developing its own generative AI model, the Samsung Gauss.
ATOM and ION.
It has also worked with customers on prior generations of semiconductors. In May 2023, Rebellions’ strategic investor, KT, integrated Atom, Rebellions’ data-center-focused AI chip, into its cloud-based neural processing units (NPU) infrastructure. Rebellions intends to collect income from Atom in the second part of the year and plans to continue producing the chip model using Samsung’s 5-nanometer manufacturing technology. Atom is intended for data centers and language models with up to 7 billion parameters, whereas Rebel is aimed at bigger language models, according to Shin.
Meanwhile, the startup’s first AI chip, Ion, announced in November 2021, is currently undergoing certification testing in the United States and has yet to sign any commercial clients. Ion is intended for edge computing, and the firm believes that one significant use case will be in financial services applications, where larger organizations developing their own hardware may utilize the chips to power stock forecasting and trading apps.
Rebellions CEO Sunghyun Park, a former quant engineer at Morgan Stanley in New York, and four co-founders launched the AI chip firm in 2020.
